Industry Insights 14 min read

What’s Inside an AI Server? A Deep PCB Cost Breakdown of NVIDIA DGX A100

This article dissects the NVIDIA DGX A100 AI server, detailing its five major hardware sections—fan module, storage, GPU board, CPU motherboard, and power module—and quantifies the PCB area and monetary value of each component to reveal where the true upgrade value lies.

Architects' Tech Alliance
Architects' Tech Alliance
Architects' Tech Alliance
What’s Inside an AI Server? A Deep PCB Cost Breakdown of NVIDIA DGX A100

Introduction

The analysis focuses on the NVIDIA DGX A100, a benchmark AI server whose chassis resembles a high‑end desktop. By reverse‑engineering the server’s printed‑circuit‑board (PCB) composition, the article aims to give readers a concrete understanding of the internal hardware and its cost implications for future upgrades.

Five Major Hardware Sections

The DGX A100 can be divided into five functional hardware groups:

Fan module (front)

Storage and front control console

GPU board tray (rear)

CPU motherboard tray (rear)

Power module (rear)

From a value perspective, the PCB contribution is grouped into three categories: the GPU board (core AI component), the CPU motherboard (common to all servers), and the remaining accessories (fans, disks, power boards).

GPU Board Group

The GPU board accounts for roughly 1.2 wan RMB per server, with 52 % of the value coming from carrier boards and 48 % from PCB layers. It consists of four sub‑parts:

GPU carrier board : 8 boards (one per GPU) using 70‑100 mm, 14‑16 layer FCBGA substrates; each valued at about 100 USD (≈650 CNY), totaling ≈5 200 CNY.

NVSwitch module : Handles high‑speed GPU‑to‑GPU communication; each unit valued at ≈30 USD (≈195 CNY), total ≈1 170 CNY.

OAM (GPU accelerator module) : One‑to‑one with each GPU; area ≈0.03 m². High‑end SXM version uses 20‑layer Ultra Low Loss CCL, priced at 12 000 CNY/m², yielding ≈2 880 CNY per server; the PCIe version is cheaper at ≈7 000 CNY/m².

UBB (Unit Baseboard) : One per server, area ≈0.30 m², 26‑layer PCB with Ultra Low Loss material, priced at 10 000 CNY/m², contributing ≈3 000 CNY.

Combined, the GPU board’s PCB area is 0.624 m² and its total PCB‑related value is ≈12 250 CNY (carrier board 6 370 CNY, PCB layer 5 880 CNY).

CPU Motherboard Group

The CPU subsystem includes:

CPU carrier boards (2 units) – each ≈100 USD, total ≈1 300 CNY.

CPU mainboard – 10‑12 layer, Low‑Loss CCL, area ≈0.38 m², priced at 3 000 CNY/m², total ≈1 140 CNY.

Functional auxiliary boards (memory, NIC, expansion, storage driver) – collectively covering 0.27 m² and valued at ≈405 CNY.

The entire CPU motherboard group occupies 0.662 m² of PCB and is valued at ≈2 845 CNY (carrier boards 46 %, mainboard 40 %, auxiliaries 14 %).

Other Accessories

Additional components such as power supplies, disks, and the front control console use 6‑10 layer FR4/Mid‑Loss PCBs, priced between 1 000‑1 500 CNY/m². Estimations based on the DGX A100 specifications give:

Power module: 6 units, each PCB area ≈0.019 m².

Disks: 8 units, each PCB area ≈0.008 m².

Front control console: 1 unit, PCB area ≈0.010 m².

These accessories contribute a modest ≈226 CNY to the overall PCB value.

Overall PCB Cost Summary

Summing all sections, the DGX A100 uses about 1.474 m² of PCB, with a total PCB‑related monetary value of ≈15 321 CNY per server. Breakdown by component type:

GPU board group: ≈12 000 CNY (≈80 % of total)

CPU motherboard group: ≈2 845 CNY (≈19 %)

Other accessories: ≈226 CNY (≈1 %)

By PCB layer classification, carrier‑board value is ≈7 670 CNY (50.1 %) and pure PCB layer value is ≈7 650 CNY (49.9 %).

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

industry insightsAI serverNVIDIA DGX A100PCB analysisGPU boardCPU motherboardhardware cost breakdown
Architects' Tech Alliance
Written by

Architects' Tech Alliance

Sharing project experiences, insights into cutting-edge architectures, focusing on cloud computing, microservices, big data, hyper-convergence, storage, data protection, artificial intelligence, industry practices and solutions.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.